Dataset Format

[kadattack]

I would like to know if this implementation of panoptic segmentation requires annotations of "bounding boxes" in the json file.
For example I used the COCO panoptic dataset 2017 where it has "bbox" information stored under "segments_info" in the json file.
Yet if I remove the the bbox from the file (panoptic_val2017.json), the training still works without any errors.

I am asking because I am trying to create a custom dataset in the format of COCO and would like to know if the bounding boxes are even required for training.

[bowenc0221]

Panoptic-DeepLab does not need bounding boxes in the json file, it only uses the segmentations.